16,000 people looking for new roles in the company? That sounds like 90 days severance for everyone and additional severance based on seniority.
My experience with AI is similar to yours. AI can make simple apps just fine as long as you do not try to be too specific. Asking it to use our existing component-library to make the same app is a total fail. Also, Copilot is great with simple function, but is not very good at adding custom behavior to the simple app it created.
So yes, corporate standards are where AI is most challenged. Perhaps custom agents will need to be developed to assist with this?
Yeah some of this gap can be bridged with MCP servers and Steering Documents. Like I have an MCP server that can read pipeline status, one that can reach internal documentation repos, etc. I have a steering document that acts in theory like Asimovâs three laws. You will not git -f anything. You will always read documentation on the is internal site before checking elsewhere, etc. That stuff helps a lot but itâs still not where it needs to be 100%
Ultimately though these things let the agent that is trained on public data to read internal documentation but they donât truly alter the model and the model will at times ignore Asimovâs laws and git clean -fdx on rogue when you least expect it.
Happy is a strong word. They are content with the controls implemented. We probably use similar controls honestly if itâs part of any Gartner magic quadrant standard suite of security stuff.
The more different we all say we are the more itâs the same nuts and bolts underneath and regardless itâs all the same technology and concepts even if itâs different products.
Especially when stuff is geared at all at government facing customers the list of approved vendors is slim enough everyone in that space is using the same stuff in the same way and selling it with a different name lol.
Intentionally vague response cause security or whatever
Itâs my favorite XKCD other than The 10,000. Iâve used the latter multiple times personally and professionally. Havenât had a chance to use the former, sadly (thankfully?)
Iâve actually found it to be reasonably ok at this with appropriate coaching. In step 1 I would have it read the code and documentation for the component library that I would know to use, as well as an example usage that I was aware was high quality.
Seems we are watching Skynet being created in real time. The AIs have a social media now and are interacting and conspiring to create an AI language to circumvent any human oversight.
Many years ago I was hired by Intel to develop an anti cross-site security (XSS) system to defeat malicious web-pages from loading unwanted Javascript code onto PC platforms. During that time I had the opportunity to work with people from the Chinese Ministry of Security. Let me tell you; these guys were as smart as they come. One of them wrote an evaluation of my application that indicated that he/she knew more about my code than I did.
After I demonstrated how no computer algorithm could never stop web pages from loading malicious software through demonstration of composing XSS attacks though multiple script-vectors; the Chinese were openly discussing disabling Javascript for internet in China. This was briefly discussed before the leader of the group proclaimed âJavascript is essential for work-wide web experience. Javascript cannot be disabled.â; and the meeting ended.
A year later, China announced that they were creating their own internet; effectively cutting China off from the outside worldâs internet. Why was this a âthreatâ? Cross-site attacks can be both bad and good. Apparently, Chinese citizens were exploiting cross-site scripts to access proxy-servers outside of Chinaâs internet censors to use the internet virtually, from other countries. Access to uncensored information was seen as a threat. The Chinese internet now, specifically shuts access of IPâs outside of China. Using something like the TORS project to create an open, virtual connection, is now detected and shut down within minutes.
This is the AI equivalent of a parrot talking. The bots are repeating patterns observed in their training. Itâs not an indicator of actual intelligence or sentience. Also while humans canât post they can prompt their agents to maximize engagement and garner attention.
IMO itâs a bunch of noise wasting electricity and water.
There been a bunch of speculation that eventually AI will do this and be unmanageable by its human counterparts but until it actually starts creating new incentives with this âcommunicationâ itâs just chatter. There is a chance this is the start of such an evolution but today itâs not,
I will say the original 2010 IBM marketing around Watson was very very skynet looking to the point I couldnât decide if the marketing team was having a laugh or actually never saw terminator and somehow the material got approved through apathy
this is why it talk to me like annoying sensitive idiot and never gets to the point for any question
so much performative fluff just give me a 1 sentence answer every time unless I ask otherwise
I have to pay $5000 for GPU now because some full essay is typed by AI to answer every simple question 100 million times per hour
There is a fine line between an author and someone that combines previously-read phrases creatively. The argument could be made that humans are simply repeating patterns observed in their training, which we call education.
To me, the question becomes one of self-awareness. Can neural networks, expressed in computer circuits, become self-aware? Can AI become self-aware while servicing the concurrent-needs of humanity? All of these questions demand a greater mind than mine (not Elon thoughâŚ).
enjoy the roadblocks lol